CNS*1994

The Annual Computational Neuroscience Meeting

July 1994, Monterey, California

CNS*1994 Abstracts

• Author: Charles H. Anderson* and David Van Essen
  Dept. of Anatomy and Neurobiology
  Washington University School of Medicine
  St. Louis, MO, 63110
  cha@shifter.wustl.edu
  vanessen@vl.wustl.edu

  Title: NEUROBIOLOGICAL REPRESENTATIONS OF INFORMATION

  Abstract: A definition for neural representations is constructed on the presumption that neurobiological systems encode joint probability density functions of analog quantities. This allows a unified framework for discussing cell responses in a variety of neural structures including the superior colliculus, visual and motor cortical areas. The formulation encompasses many aspects of standard neural network modeling, while imposing a rigorous framework for analysis and understanding. Biological systems can be seen as displaying high variability from one perspective, while performing robust and consistent computations at another.

  ---

• Author: J.L. Armony, and J.E. LeDoux
  Center for Neural Science
  New York University,
  4 Washington Place,
  New York, NY l0003
  
  D. Servan-Schreiber
  Department of Psychiatry
  University of Pittsburgh School of Medicine
  3811 O'Hara Street
  Pittsburgh, PA 15213

  Title: NEURAL SYSTEM OF FEAR CONDITIONING: A CONNECTIONIST MODEL

  Abstract: We developed a connectionist model of the thalamo-cortico-amygdala network that mediates conditioned fear responses to auditory stimuli. The relevant neural structures are represented as modules of non-linear units with lateral inhibition, connected by feedforward parallel pathways of excitatory weights. The network is trained using a Hebb-type learning rule. The model accounts for the behavior and . frequency receptive-field changes observed empirically during conditioning and provides a new approach to the study of emotional information processing.

  ---

• Author: Thomas J. Anastasio
  University of Illinois
  Beckman Institute

  Title: Recurrent backpropagation models of the Vestibulo-Ocular Reflex provide experimentally testable predictions

  Abstract: Previous, static backpropagation models of the vestibulo-oculomotor system were able to capture the distributed aspects of eye-movement command representation by brainstem neurons. However, these models do not readily offer testable predictions. More recently, recurrent backpropagation models have been used to study the dynamic and nonlinear features of the vestibulo-ocular reflex (VOR). The dynamic models make clear predictions concerning the behavior of VOR neurons following lesions. Some of the predictions from the recurrent backpropagation models differ in critical ways from those derived from analytical models of VOR. The testability of the recurrent models encourages a continued dialog between theory and experiment.

  ---

• Author: Evyatar Av-Ron*
  Groupe de Bioinformatique, URA 686 - Ecole Normale Superieure, 46
  rue D'Ulm, 75230 Paris cedex 05, France

  Hanna Parnas
  Department of Neurobiology, Institute of Life Sciences, Hebrew
  University, Jerusalem 91904, Israel

  Lee A. Segel
  Department of Applied Mathematics and Computer Science,
  Weizmann Institute of Science, Rehovot 76100, Israel

  Title: MODELING THE BURSTING INTERNEURONS OF THE LOBSTER CARDIAC GANGLION

  Abstract: The lobster cardiac ganglion consists of four interneurons and five motorneurons that stimulate the lobster heart. The interneurons exhibit bursting behavior of varied durations and frequencies. They consist of at least one pacemaker (endogenous burster) and possibly conditional bursters. A simple biophysical model is used to describe the four interneurons.

  Based on experimental results, the parameters of the model are chosen to fit the description of the slow calcium wave observed in the soma. This calcium wave is considered a driver potential for the bursting behavior. In addition, Hodgkin-Huxley sodium and potassium currents are incorporated. The full model exhibits the action potentials riding on the slow wave. The rates of calcium influx and removal control the durations of bursting and quiescence while the maximal sodium conductance controls the frequency during the burst.

  ---

• Author: H. Axelrad,
  J. Laine, B. Berthie, A. Crivat & M.E. Marc
  Laboratoire de Neurophysiologie,
  Faculte de Medecine PitieSalpetriere,
  91 bd de l'H6pital, 75634 Paris Cedex 13, France

  Title: SOME NEW FINDINGS ABOUT THE NEURONAL TYPES OF THE CEREBELLAR CORTEX AND SOME HYPOTHESIS ABOUT POSSIBLE ADDITIONS TO THE WIRING DIAGRAM OF THIS STRUCTURE.

  Abstract: Our views about the elementary components and the interrelating circuitry of the corticocerebellar network have not changed since Ramon y. Cajal, whose superb morphological work was functionally confirmed by Eccles and collaborators. Four recent findings, among which the discovery of three new cell types, have been made, on Golgi impregnated material, and add more complexity to what was considered to be a rather simple structure.

  1) The axon of the fusiform cell of Lugaro has now been shown to terminate in the molecular layer; 2) Another large, but globular, neuron of the granular layer sends a profusely branched axon to the molecular layer; 3) The candelabrum cell, located inside the ganglionic layer, also has an axon that terminates in a peculiar fashion in the molecular layer; 4) The unipolar brush cell, found mostly in the granular layer of the vestibulocerebellum, has a specific monodendritic appearance and a contorted axon, with rosette excrescences, that stays confined into the same layer.

  These new findings will be described, including 3D reconstructions, and the possible additions to the wiring diagram of this structure discussed.

  ---

• Author: H. Axelrad, O. Chatelain, and C. Bernard
  Laboratoire de Neurophysiologie,
  Faculte de Medecine PitieSalpetriere,
  91 bd de l'Hopital, 75634 Paris Cedex 13, FRANCE.

  Title: A SIMULATION MODEL OF THE MOSSY FIBERS AND GRANULE CELLS OF THE CEREBELL~R CORTEX

  Abstract: The information brought to the cerebellar cortex by mossy fibers is relayed to the Purkinje cells and other interneurons dendritic trees by means of the granule cells and their axons, which branch in the molecular layer as parallel fibers. A recent realistic simulation of these parallel fibers has shown that the properties of travelling volleys imply that the Purkinje cells act as coincidence detectors (C. Bernard & H. Axelrad, Brain Research, 1991, 565: 195-208) a finding that was confirmed by a theoretical anatomical analysis (J. Meek, Neuroscience, 1992, 48: 249-283).

  We extend here our former parallel fiber simulation study by building a model of a restricted volume of the granular layer in which mossy fibers and granule cells are represented. Realistic morphological and physiological data are used and the model is constructed so as to allow further additions, such as Golgi cells, and interaction with the parallel fiber model. The morphological and physiological elements embedded in the model will be extensively described as well as the rules governing the mossy fiber-granule cell interactions. First results of simulations, under constrained conditions, will be discussed.

  ---

• Author: Wyeth Bair1,2, Christof Koch 1, William Newsome 3, and Kenneth
  Britten 4

  1 Computation and Neural Systems Program
  California Institute of Technology
  Pasadena, CA 91125

  2 California Institute of Technology
  139-74
  Pasadena, CA 91125

  3 Department of Neurobiology
  Standford University School of Medicine
  Stanford, CA 94305

  4 Center for Neuroscience
  University of California Davis
  Davis, CA 94516

  Title: RELIABLE TEMPORAL MODULATION IN NEURONAL SPIKE TRAINS IN AREA MT OF AWAKE MACAQUE MONKEY

  Abstract: We analyzed the repeatability of temporal structure in spike trains recorded from area MT in behaving rhesus monkeys which viewed a dynamic random dot stimulus (Newsome, Britten & Movshon, 1989). We find a surprising degree of regularity in the temporal modulation of the response of many cells--the most reliable cell will fire a few isolated action potentials or show a period of elevated firing which begins and ends fixed in time relative to stimulus onset with a standard deviation of less than 4 msec in recording periods which may last many minutes to hours. Spike count during these short periods of elevated firing rate is more reliable, i.e. has a lower variance to mean ratio, on average than the spike count taken over the entire 2 sec stimulus. Onset and offset of high firing periods can occur with very similar and fast time constants. Autocorrelation analysis reveals that deviations from the mean response are rarely substantially correlated beyond 100 msec, and in half of cell no correlation exists between consecutive interspike intervals beyond that predicted from the time-varying mean firing rate. Reliable temporal modulation intervals beyond that predicted from the time-varying mean firing rate. Reliable temporal modulation disappears or diminishes for completely coherent motion stimuli, and we propose an experiment to determine whether this is due to saturation in firming rate or due to a qualitative change in the way MT cells respond to the stimulus.

  ---

• Author: Dimitrios Bairaktaris,
  Dept. of Computing Science,
  University of Stirling,
  Stirling FK9 4LA,
  Scotland, UK

  Title: LOCALISED NEURONAL ASSEMBLIES WITH RECURRENT SYNAPSES ENABLE GLOBAL TEMPORAL SYNCHRONISATION

  Abstract: Neuronal assemblies with recurrent synapses can be found in area CA3 of the hippocampal formation. It has been suggested that such neuronal formations play an important role in temporal information processing. The work presented in this paper describes a modular network architecture comprising localised recurrent networks which can perform global temporal synchronisation. A situation where temporal syn~hronisation is required, arises in human Short Term Memory research where individual features of a single memory object are encoded at different points within a very small window of time and are then dynamically bind together to form the representation of the object. This form of dynamic binding can be achieved by combining existing activation maintenance and temporal synchronisation computational techniques.

  ---

• Author: Bill Baird* 1, Todd Troyer 2 and Frank Eeckman 3

  1 Department of Mathematics
  U.C. Berkeley
  Berkeley, CA 94720

  2 Dept. of Phys.,
  U.C. San Francisco
  513 Parnassus Avenue
  San Francisco, CA 94143

  3 Frank Eeckman
  Lawrence Livermore National Laboratory
  P).O. Box 808 (L-0270)
  Livermore, CA 94550

  Title: ATTENTION AS SELECTIVE SYNCHRONIZATION OF OSCILLATING CORTICAL SENSORY AND MOTOR ASSOCIATIVE MEMORIES

  Abstract: We show how a sensory/motor network architecture, constructed from recurrently connected oscillatory associative memory network modules, can employ selective "attentional" control of synchronization to direct the flow of communication and computation within the architecture to solve a grammatical inference problem.

  In this architecture, oscillation amplitude codes the information content or activity of a module (unit), whereas phase and frequency are used to "softwire" the network. Only synchronized modules communicate by exchanging amplitude information; the activity of non-resonating modules contributes chaotic crosstalk noise.

  Attentional control is modeled as a special subset of the modules with ouputs which affect the resonant frequencies of other modules. They learn to control synchrony among these modules and direct the flow of computation (attention) to effect transitions between subsections of a large automaton which the system learns to emulate. The internal crosstalk chaos is used to drive the required random transitions of the system.

  ---

• Author: Pierre Baldi* 1 and James M. Bower 2

  1 Division of Biology 216-76
  California Institute of Technology
  Pasadena, CA 91125

  2 Division of Biology 216-76
  California Institute of Technology
  Pasadena, CA 91125

  Title: COMPUTATIONAL APPROACHES TO EARLY OLFACTION

  Abstract: We have recently been interested in developing a better understanding of early olfaction, from both a molecular and a computational point of view. At the molecular level, we have used Hidden Markov Models to analyze the amino acid sequences of the large family of recently described putative G-protein related olfactory receptors. When the primary structures of these proteins are compared to those of other brain G-coupled receptors, the putative olfactory receptors are found to be the most variable. Further, a high degree of randomness appears to be present in the amino acid structures of these proteins. The diversity seen in the arnino acid sequences of the putative olfactory receptors seems to be perfectly consistent with the overall physiological and anatomical properties of the olfactory system. For example, this system is known for its extremely broadly tuned neuronal responses as well as its diffuse and distributed pattern of inter neuronal connectivity. Further, the olfactory receptor neurons turn over rapidly (1/2 life 60 days) throughout the life time of the animal. We have recently developed several mathematical models with which to explore the potential significance of these features of the olfactory system for information processing. These models share many of the known features of the projections from olfactory receptors to the mitral cells of the olfactory bulb, but differ in the extent of topography in the projections of receptors of different types to the bulb. Using statistical techniques, we have contrasted the likely ability of each model to adequately represent olfactory stimulus space. The results suggest that the highly distributed almost random patterns of cellular response and connectivity may be the most efficient means to assure an adequate sample of olfactory stimulus space. In this context, the models also provide a means to interpret the tremendous diversity in the molecular structure of the olfactory receptors we have observed.

  ---

• Author: Bell A.J., Mainen Z.F. & Sejnowski T.J.ORAL

  Computational Neurobiology Laboratory
  Salk Institute
  La Jolla, CA

  Title: 'BALANCING' OF CONDUCTANCES MAY EXPLAIN IRREGULARITY OF CORTICAL SPIKING.

  Abstract: How may synaptic and voltage-dependent conductances be adjusted in models to achieve spiking as noisy as that seen in cortical neurons? We identify five factors contributing to inter-spike-interval (ISI) irregularity: the mean and variance of the input current, the instananeous membrane resistance, the degree of repolarisation after spiking and bistabilities in the membrane dynamics. Crucially, by balancing excitation and inhibition so the cell is typically around threshold, we are able to achieve ISI coefficients of variation of around 1 in single compartment models. Our simulations suggest that the currents entering a neuron are 'balanced' to achieve maximum sensitivity to inputs.

  ---

• Author: Biederman* 1, Jozsef Fiser 2, and Eric E. Cooper 3

  1 Hedco Neuroscience Building
  University of Southern California
  Los Angeles, CA 90089-2520

  2 Center for Neural Engineering, Image Understanding Lab Hedco
  Neuroscience Building University of Southern California Los Angeles,
  CA 90089-2520

  3 Deparrtment of Psychology University of Minnesota Elliott Hall
  Minneapolis, MN 55455

  Title: TEST OF A TWO-LAYER NETWORK AS A MODEL OF HUMAN ENTRY-LEVEL OBJECT RECOGNITION

  Abstract: A number of recent models of shape recognition assume that the outputs of a lattice of early (V1) spatial filters are mapped directly onto an object representation layer. Although such two-layer networks may be appropriate for face recognition or visually guided motor behavior, a near optimum version of such a model failed to generate the qualitative characteristics of human object recognition data. Hidden layers that represent viewpoint-invariant properties of contours may be required for modeling human object recognition.

  ---

• Author: Kenneth I. Blum* and Marco A. P. Idiart
  Center For Complex Systems,
  Dept. of Biology, and Dept. of Physics, Brandeis University,
  Waltham, MA 02254-3110

  Title: A THEORETICAL FRAMEWORK FOR QUANTAL ANALYSIS AND ITS APPLICATION TO LONG TERM POTENTIATION

  Abstract: We have constructed a mathematical framework for quantal analysis of central synaptic transmission that takes both multiple synapses and postsynaptic fluctuations into account. For each instant in time we calculate the distribution of postsynaptic currents produced by synaptic events. We have used this framework to determine the shape of distributions both close to and far from receptor saturation. Our approach allows the effects of changing both presynaptic and postsynaptic parameters to be conveniently assayed.

  ---

• Author: Edward K. Blum* 1, Qingnan Li 2, Stephen C. J. Hyland 3, and Patrick K. Leung 4

  1 Department of Biomedical Engineering
  2 Department of Mathematics
  3 Department of Computer Science
  4 Department of Biomedical Engineering
  University of Southern California
  Los Angeles, California 90089

  Title: BIONNIC: AN EFFICIENT AND FLEXIBLE INTEGRATOR FOR BIOLOGICAL NEURAL NETWORK SIMULATORS

  Abstract: We present the Biological Neural Network Integrator Collection, BIONNIC, a stable and efficient integrator for computing solutions to large systems of ordinary differential equations obtained from compartmental modeling of networks of neurons, each neuron having an arbitrarily branched tree structure. BIONNIC is a portable and reusable library of C subroutines which differs from many general purpose integrators (LSODE, IVPAG) by permitting multiple calls for different sets of equations to be intermixed, and by dynamically allocating memory. This expedites the efficient implemention of parallel simulations of biological neural networks. In addition to fixed time step modes, BIONNIC has variable step, variable order (VSVO) backward differentiation formulae (BDF), which are stiffly stable lor orders 1 to 6, combined with an O(n) time complexity algorithm due to Parter, for solving a linear system of algebraic equations derived from implicit methods.

  Currently, BIONNIC has been implemented in the new version of our simulator, Cajal-V4.0, on a Sun Sparc 10. In VSVO mode, for a given accuracy, the best combination of order and step size is calculated after each integration step, so that the dynamic activity of each neuron will be taken into account to shorten the computation time. Preliminary simulation results have shown BIONNIC to be accurate, fast, and portable. BIONNIC will be available through anonymous ftp, as of July 1, 1994.

  ---

• Author: Charlotte Boettiger* and Gwen Jacobs
  Dept of Molecular and Cell Biology Neurobiology Division University
  of California 195 LSA Berkeley, CA 94720

  Title: A QUANTITATIVE ANALYSIS OF POST-EMBRYONIC DEVELOPMENTAL CHANGES IN THE CRICKET CERCAL SENSORY SYSTEM'S FUNCTIONAL MAP OF WIND DIRECTION.

  Abstract: Using methods of computer aided reconstruction and morphological analysis developed in our lab, we have analyzed the structures of a representative sample of indentified neurons in the afferent projection of the cricket cercal sensory system's functional map of wind direction, their spatial relationships to one another, and the map as a whole, at 30%, 50%, 70%, and 90% of post-embryonic development. Each the cells studied has a characteristic gestalt which is conserved through development, but at earlier stages the arborizations are less elaborated and the architecture of the cell's original targeting can be seen. The global structure of the map exists in it's ultimate form by 30% of post-embryonic development, but as cells elaborate, they overlap more with those having related functional properties and the local organization of the map becomes more continuous.

  ---

• Author: Eric Boussard* and Jean-Francois Vibert

  B3E, Inserm U263
  Faculte de Medecine Saint-Antoine
  27 rue Chaligny
  75571 Paris cedex 12

  Title: A MODEL OF RETINA INCLUDING DOPAMINERGIC NEUROMODULATION

  Abstract: The fovea of a mammal retina was simulated with detailed biological properties to study the images preprocessing. The direct pathway (photoreceptors, bipolar and ganglion cells), the horizontal units, and the D-amacrine cells were simulated. The computer program simulated particularly the gapjunctions between horizontal cells and their dopaminergic neuromodulation. This retina was able to reproduce biological observations, contour extraction with Mach effect, adaptation to brightness, progressive disappearance of non moving images, reversed post-image, and optic illusion (Hermann grid). The simulations showed that dopaminergic amacrine cells were necessary to ensure adaptation to local brightness, and to keep a good dynamic of brightness response.

  ---

• Author: Dean Buonomano* and Michael Merzenich

  Keck Center for Integrative Neuroscience
  University of California
  San Francisco
  P.O. Box 0732
  San Francisco, CA 94143

  Title: A CORTICAL MODEL OF TEMPORAL INFORMATION PROCESSING BASED ON PAIRED-PULSE FACILITATION AND SLOW IPSPS.

  Abstract: The processing of sensory information from afferent pathways relies not only on the spatial patterning of the inputs, but also on the temporal relationship between the inputs. How the nervous system processes temporal information is unclear. Various neuronal properties, such as paired-pulse facilitation and slow IPSPs, exhibit timedependent properties, but their role in neural information processing is unknown. In order to address whether such time-dependent properties may underlie temporal processing we have developed a continuous-time artificial neural network based on integrate-and-fire elements that incorporate paired-pulse facilitation and slow IPSPs. By incorporating these elements into a circuit inspired by neocortical connectivity, we demonstrate that the network is able to discriminate different temporal patterns. Generalization emerges spontaneously. We propose that paired-pulse facilitation and slow IPSPs play an important role in information processing by permitting a network to encode temporal and sequential information by altering the firing probability of neurons in a time-dependent manner.

  ---

• Author: Pierre-Yves Burgi* and Norberto M.Grzywacz

  The Smith-Kettlewell Eye Research Institute,
  2232 Webster Street, San Francisco, CA 94115.

  Title: HEBBIAN PROCESSES AND SPONTANEOUS WAVES OF ACTIVITY COULD LEAD TO THE EMERGENCE OF COMPLEX RETINAL RECEPTIVE FIELDS

  Abstract: Center-surround and orientation selective cells can emerge from spontaneous (uncorrelated) activity, as previously demonstrated by Linsker's multi-layer Hebbian model. We wanted to test whether the spontaneous waves of activity present in a single (inner plexiform) layer of developing retinas could account for the emergence of such selectivities in ganglion cells. An eigenfunction analysis of a Linsker-like model with waves at the input indicated that only one synaptic layer is required for the emergence of these selectivities. The analysis also showed that waves favor symmetry, an effect that is reduced when the angle formed by the orientation of the wave and presynaptic dendrites is taken into account. Supported by the SNFSR (8220-37180) and NEI (EY08921).

  ---

• Author: Matteo Carandini 1 and David J. Heeger 2

  1 Center for Neural Science
  New York University
  6 Washington Place, Rm 809
  New York, New York 10003
  (212) 998 7898

  2 Psychology Department
  Stanford University
  Building 420
  Stanford, California 94305
  (415) 723 4048

  Title: SUMMATION AND DIVISION IN V1 SIMPLE CELLS

  Abstract: We model simple cells in the primary visual cortex as performing a weighted sum of the image intensities over space and time, followed by mutual divisive suppression. The cell membrane performs both operations: the input currents get added and are divided by the overall membrane conductance. We show how current and conductance can be completely decoupled if the synaptic inputs are segregated in two groups, one of them showing a complementary arrangement of excitation and inihibition. The model accurately predicts the responses of monkey Vl simple cells to sinusoidal gratings varying in contrast, orientation and velocity.

  ---

• Author: Nicholas T. Carnevale, Kenneth Y. Tsail, Brenda J. Claiborne, and
  Thomas H. Brown
  Nicholas T. Carnevale
  Department of Psychology
  Yale University
  P.O. Box 208205
  New Haven, CT 06520-8205

  Title: QUALITATIVE ELECTROTONIC COMPARISON OF THREE CLASSES OF HIPPOCAMPAL NEURONS IN THE RAT

  Abstract: We describe the algorithms for an efficient and rapid transformation that maps neuronal anatomy into electrotonic space. This transformation reveals the spatiotemporal dynamics of electrical signaling in nerve cells in a way that facilitates rapid intuition of the functional consequences of neuronal architecture. We present the results of using this transformation in a qualitative, preliminary investigation of the comparative electrotonic structure of three classes of hippocampal neurons in the rat (CAl and CA3 pyramidal cells, and dentate granule cells). This revealed unanticipated similarities and differences within and between these cell classes that may be of importance to the function of hippocampal circuitry.

  ---

• Author: C. E. Carr

  University of Maryland
  College Park, MD 20742

  Title: THE DEVELOPMENT OF NUCLEUS LAMINARIS IN THE BARN OWL

  Abstract: The barn owl uses interaural time differences to localize sound in azimuth. Sensitivity to these interaural time differences (ITD) arises in the brainstem nucleus laminaris. Maps of ITD are formed in the dorso-ventral dimension of the nucleus laminaris by interdigitating axons from the ipsi- and contralateral magnocellular cochlear nuclei. The amount of delay mapped in the nucleus laminaris depends on the length and conduction velocity of the magnocellular axons.

  ---

• Author: Hung-Jen Chang* and Walter J. Freeman

  Division of Neurobiology
  Department of Molecular & Cell Biology
  University of California
  Berkeley, CA 94720-3200

  Title: PARAMETER OPTIMIZATION IN AN OLFACTORY NEURAL SYSTEM

  Abstract: We study an olfactory system that is characterized physiologically by spatiotemporal connections among neural ensembles and external inputs. The behavior of each ensemble is modeled by a 2nd order ODE relating the aggregate activation of cells to system parameters. Parameter optimization rules that lead to minimizing the distance between the actual outputs of the model and the EEG waves from experimental data are derived by either the method of error propagation or calculus of variations. The existence and uniqueness of the equation set are also proved to assure the propriety of parameter adaptation. Subsets of the entire system are simulated by computer programs. Numerical results support the mathematical analyses and the parameters are thus optimized.

  ---

• Author: Andrew Chapman and Suzanna Becker*

  Department of Psychology
  McMaster University Hamilton
  Ontario, Canada, L8S 4K1

  Title: MODEL SYNAPSES WITH FREQUENCY POTENTIATION CHARACTERISTICS CAN COOPERATIVELY ENHANCE HEBBIAN LEARNING

  Abstract: Frequency potentiation, a short-term form of plasticity, is an enhancement in the amplitude of neuronal responses to each pulse in a train of stimulation pulses which occurs when the pulses are delivered within a certain frequency range. In the model, theta-frequency input from the subiculum via synapses with frequency potentiation characteristics cooperatively enhances Hebbian learning in pyriform (olfactory) cortex efferents to the entorhinal cortex, particularly when the inputs are phase-locked. This effect is further enhanced when inhibitory neurons are added at the entorhinal layer. This is the first report of a network model in which the possible computational functions of synapses with frequency potentiation characteristics are explored.

  ---

• Author: Heather Clague*, Cooper Roddey, Frederic Theunissen and John Miller.

  Dept. of Molecular and Cell Biology
  Neurobiology Division
  195 LSA. Berkeley, CA 94720.

  Title: THE EFFECT OF ADAPTATION ON THE QUALITY OF CODING OT STIMULI FOR AFFERENTS AND INTERNEURONS IN THE CRICKET SENSORY SYSTEM

  Abstract: We demonstrate that some neurons of the cricket cercal sensory system are capable o substantial adaptation; their steady state coding accuracies remains the same for a wide range o stimulus powers. The process of adaptation has a rapid and a slow component, and the accuracy of coding changes during the adaptation process. In adapting to high power levels, the overal accuracy measured in baud actually decreased as the spike rate decreased. However, the decrease in accuracy was smaller than the corresponding decrease in spike rate, resulting a larger net accuracy (or "infonnation content") per spike in adapted cells.

  ---

• Author: M. Elizabeth Corey

  Department of Neurobiology and Physiology and Program in
  Computer Science
  2153 North Campus Drive
  Northwestern University
  Evanston, Illinois 60208

  Title: PARALLEL CONTROL MECHANISMS IN COCHLEAR MECHANICS

  Abstract: A cochlea contains nearly 12,000 sensory receptor cells that change shape in response to receptor excitation. We modeled the cell with dual inputs (mechanical and synaptic excitation) and used physiologically-based equations for receptor transduction, basolateral membrane potential, and somatic length changes to determine responses. We incorporated the cell's shape changes into the boundary conditions for cochlear fluid mechanics and solved using a finite-difference time-domain solution strategy.

  A parameter estimation scheme is used to define unknown parameters. Estimation is guided by stability measures (Liaponuv exponents, sensitivity derivatives and the Gaussian curvature), the occurrence and location of bifurcations, and the size, fractal dimension and power spectrum of attractors. Some results as well as the advantages and limitations of using such a scheme are described.

  ---

• Author: E. De Schutter 1 and J.M. Bower 2

  1 Born Bunge Foundation
  University of Antwerp - UIA, B2610
  Antwerp, Belgium

  2 Div. of Biology 216-76
  California Institute of Technology
  Pasadena, CA 91125

  Title: SHORT-TERM INTERACTIONS BETWEEN THE COMPLEX SPIKE AND GRANULE CELL INPUTS IN THE PURKINJE CELL

  Abstract: We used a detailed compartmental model of a Purkinje cell to investigate how a climbing flber input might change the response to subsequent granule cell inputs. We have previously shown that responses to small synchronous granule cell inputs are amplified by the same dendritic Ca2+ channels which are also activated by the complex spike. A complex spike fired up to 160 ms before the synchronous input had no effect on somatic EPSP amplitude. At shorter intervals, there was first a small potentiation, followed by a depression, followed by a large potentiation. This may explain the conflicting results reported from extracellular recordings in vivo.

  ---

• Author: Alain Destexhe, Zachary F. Mainen, and Terrence J. Sejnowski

  The Howard Hughes Medical Institute and
  The Salk Institute,
  Computational Neurobiology Laboratory,
  10010 North Torrey Pines Road, La Jolla, CA-92037

  Title: ELEMENTARY KINETICS PROVIDE EFFICIENT MODELS OF SYNAPTIC TRANSMISSION AND NEUROMODULATION

  Abstract: Markov kinetic models allow the dynamics of ionic currents to be related directly to changes in the conformational states of underlying channel proteins and the behavior of the channels to be represented in the same formalism as that used to describe biochemical processes. We use such an approach to explore models of both transmitter-gated and second messenger-gated synaptic channels. We show that elementary Markov schemes capture the essential properties of whole-cell recorded synaptic currents. The simplicity of these models makes them very efficient tools for simulating synaptic currents.

  ---

• Author: Frank H. Eeckman* 1, Richard M. Durbin2, Jean-Thierry Mieg3, and Martin Lades 1

  1 Institute for Scientific Computing Researeh
  Lawrence Livermore National Laborator;y
  Livermore, CA 94551

  2 Laboratory for Molecular Biology
  Medical Research Council,
  Cambridge, England

  3 Centre de Recherehe en Biologie Moleculaire
  CNRS, Montpellier, Franee;

  Title: BRAINACE: A DATABASE-HYPERMEDIA TOOL FOR NEUROSCIENTISTS

  Abstract: BrainAce is an object-oriented database based on ACEDB, a database system originally developed for the C. elegans genome project (JTM and RD). Although originally written for Unix/X windows, a Macintosh interface is available (FE and RD) that is functionally identical to the Unix version.

  ACEDB consists of a core data manager and specific application code. Data are stored in objects that are organized into classes. The objects have extendable structure, so that arbitrarily large amounts of information can be stored in them. That information may include annotations of various sorts, comments, and cross-references. The schema specifying data structures can be extended during the lifetime of a database. Brainace includes neuroscience-specific application code and multimedia extensions not found in the genome versions. There is a browser mode and a general search facility to provide maximal flexibility. In addition, displays can be output in Postscript for laser printing, or as plain text for transfer to other sources.

  ---

• Author: John G. Elias and David P. M. Northmore

  University of Delaware
  Newark, DE 19716

  Title: VLSI NEUROMORPHS: BUILDING BLOCKS FOR NEURAL CIRCUITS

  Abstract: Our VLSI neuromorphs consist of (a) extensive dendritic trees with hyperpolarizing, depolarizing, and shunting synapses which can receive large numbers of pulsatile inputs, and (b) integrate-and-spike somas. Programmable tree and soma dynamics and activity-dependent soma threshold, in addition to flexible synaptic connections enables a single neuromorph to perform a variety of spike processing functions, such as discriminating spatio-temporal patterns of input spikes. Adjustment of dendritic dynamics and soma parameters allows neuromorphs to be tuned over very wide temporal frequency ranges.

  ---

• Author: Erik Fransn* and Anders Lansner

  SANS - Studies of Artificial Neural Systems
  Dept. of Numerical Analysis and Computing Science
  Royal Institute of Technology
  S-100 44 Stockholm, Sweden

  Title: LOW SPIKING RATES IN A NETWORK WITH OVERLAPPING ASSEMBLIES

  Abstract: In this study we show that low rate sustained after-activity can be obtained in a simulated network of overlapping assemblies. The low rate is achieved by assuming that the synapses in the network are of a saturating type. After-activity is produced when the application of a monoamine neuromodulator is simulated. The network gives pattern completion of incomplete input and shows noise tolerance. Despite the overlap, the activity of one assemby does not spread to others. The time to reach a full pattern, the "reaction time", is only, 40-100 ms. When arts of two patterns are presented simultaneously a rivalry process can lead to full completion of one pattern and supression of the other

  ---

• Author: Walter Freeman* 1, Richard X. Tang, 2 and David Gramham-Squires 1

  1 Department of Molecular and Cell Biology
  Division of Neurobiology, 129 LSA
  University of California
  Berkeley, CA 94720

  2 Biophysics Graduate Group, 129 LSA
  University of California
  Berkeley, CA 94720

  Title: PHASE ANALYSIS OF MULTICHANNEL EEGs FROM PREPYRIFORM CORTEX IN RABBIT

  Abstract: EEGs are recorded simultaneously from an 8x8 array of electrodes on the rabbit's prepyriform cortex. Fast Fourier Transform(FFT) and nonlinear fitting techniques yield 64 phase values. The results show that a stable phase pattern recurs with each respiratory burst but not during the interburst periods. Spatial phase standard deviations are calculated before and after the median phase pattern is removed. A reversal in the phase standard deviation of the 40-80Hz bandpass filtered data is found. We postulate that this reversal confirms the hypothesis that the cortex is driven by the olfactory bulb. No centrifugal or endogenous origin of a phase pattern is revealed.

  ---

• Author: Walter J. Freeman* and John M. Barrie

  Department of Molecular and Cell Biology Neurobiology Division
  129 Life Sciences Addition
  University of California at Berke1ey
  Berkeley, CA 94720

  Title: PERCEPTUAL TOPOGRAPHY: SPATIOTEMPORAL ANALYSIS OF PREPYRIFONN, VISUAL, AUDITORY, AND SOMESTHETIC EEGS IN
  PERCEPTION BY TRAINED RABBITS

  Abstract: We tested the hypothesis that perceptually-related spatio-temporal patterns would be found in the EEG of neocortex. Arrays of 64 electrodes were fixed onto the epidural surfaces of the prepyriform, visual, auditory, and somesthetic cortices of 18 rabbits. After recovery, the subjects were trained to respond to simple conditioned stimuli in a classical aversive paradigm. The 64 EEG traces were recorded in 6sec trials, stored on disk, segmented, and decomposed by FFT, PCA, or RMS analysis. Spatial pattern differences were determined by a Euclidean distance and assigned a probability value. Differences between CS+ and CS- segments were found in narrow time windows post-stimulus. We concluded that the neural processes for perception are similar for sensory paleo- and neocortex. In all of these systems, perceptions are constructed by chaotic dynamics.

  ---

• Author: Walter J. Freeman* 1, Richard X. Tang 2, and Kouichi Sugitab 3

  1 Department of Molecular and Cell Biology
  Division of Neurobiology, 129 LSA
  University of California
  Berkeley, CA 94720

  2 Biophysics Graduate Group, 129 LSA
  University of California
  Berkeley, CA 94720

  3 P.O.Box 11165
  University of California
  Santa Barbara, CA 93107

  Title: USING FICTIVE POWER AS EVENT MARKER IN EEG DATA PROCESSINGL

  Abstract: We are presently implementing the calculation of fictive powers and using the program to process our rabbits' visual cortex EEG data. We show that the fictive powers are useful in understanding cortex EEG recordings. They are especially useful in identifying oscillatory events. We postulate that they can be used as markers in EEG data processing.

  ---

• Author: Kohyu Fukunishi*, Nobuyuki Murai,Tsuyoshi Miyashita, and Ryo Tokiok

  Advanced Research Laboratory, Hitachi, Ltd.
  Hatoyama, Saitama, 350-03, Japan

  Title: EMPIRICAL NEURAL NETWORKS AND DYNAMIC CHARACTERISTICS OF THE GUINEA PIG AUDITORY CORTEX BY OPTICAL IMAGING

  Abstract: We analyzed stimulus induced evoked responses obtained by optic imaging with dye (RH 795) in primary auditory area (A1) of the guinea pig auditory cortex. All signal (128-channel) were recorded with a spatial resolution of 130-217 um and with a time resolution of 10 kHz. Click and tone burst were applied as the sound stimuli. The dynamic spatio temporal response to click and the stable response to tone were observed and discussed. Neural oscillations about 20-50 Hz in the auditory responses were found. Cortical neural binding structure in the tonotopic organization was estimated by pattern time series analysis.

  ---

• Author: Fabrizio Gabbiani* and Christof Koch

  Division of Biology, 139-74
  Caltech, Pasadena CA 91125

  Title: ESTIMATION OF TIME-VARYING SIGNALS ENCODED IN NEURONAL SPIKE TRAINS

  Abstract: We study the linear decoding of time-varying signals from neuronal spike trains by means of analytic and computer simulation methods. Decoding algorithms were first applied by Bialek et al. (Science, vol. 252, pp. 1854-1857) in experiments on the visual system of the fly. We consider a simple model of neurons encoding time-varying stimuli in Poisson spike trains. Under some general assumptions, we are able to derive an explicit formula for the optimal decoding filter. Our results show that it depends in a non-trivial way on the statistics of the stimulus (power spectral density) as well as on the firing rate of the neurons. We examine more complex models by means of computer simulations.

  ---

• Author: Kurt A. E. Geitz* 1 and Allan Gottschalk 2

  1 M.S. Department of Bioengineering
  University of Pennsylvania
  Philadelphia, Pennsylvania 19104

  2 Department of Anesthesia
  Hospital of the University of Pennsylvania
  3400 Spruce Street Philadelphia, Pennsylvania 19104

  Title: THE ROLE OF FEEDBACK IN RESPIRATORY AMPLITUDE AND FREQUENCY MODULATION

  Abstract: To establish its contribution to modulating the pattern of respirationt the role of sensory feedback to the central respiratory pattern generator (CRPG) is investigated analytically and computationally. The limit cycle behavior of the isolated CRPG allows a generic model of the closed system to be reduced to one that depends on the amplitude of respiration and the phase difference between the CRPG and the lungs We find that the hyperbolic relationship between respiratory volume and period seen in vivo critically depends upon the nonlinear threshold to mechanical feedback Computational experiments using a network of neurons model and a pacemaker model confirm the independence of the result from the source of the limit cycle